CN109918615A - A kind of multi-modal recommended method and device - Google Patents
A kind of multi-modal recommended method and device Download PDFInfo
- Publication number
- CN109918615A CN109918615A CN201811587985.7A CN201811587985A CN109918615A CN 109918615 A CN109918615 A CN 109918615A CN 201811587985 A CN201811587985 A CN 201811587985A CN 109918615 A CN109918615 A CN 109918615A
- Authority
- CN
- China
- Prior art keywords
- tensor
- matrix
- user
- article
- modal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The present invention relates to automatic recommender system field more particularly to a kind of multi-modal recommended methods, comprising: the attribute information based on user constructs multi-modal user's space；Attribute information based on article constructs multi-modal article space；Based on the multi-modal user's space and the multi-modal article space, the scoring tensor between user and article is obtained；Non-negative orthogonal tensor singular value decomposition is carried out to the scoring tensor, obtains the approximate tensor of the scoring tensor；Based on the appraisal result in the approximate tensor, recommend adaptable article for user, and then use non-negative restriction, guarantees that the recommendation score result finally obtained is all nonnegative value, improve recommendation effect.
Description
Technical field
The present invention relates to automatic recommender system field more particularly to a kind of multi-modal recommended method and devices.
Background technique
Nowadays the appearance of internet brings a large amount of data information, meets user the information age the needs of, still, with
The rapid development bring network information amount of network increase substantially so that user can not therefrom obtain when facing bulk information
The part information actually useful to oneself is obtained, the service efficiency of information is reduced instead, to cause asking for information overload
Topic.
In order to solve the problems, such as information overload, use recommender system that can quickly recommend the information for being suitble to user for user,
The recommender system passes through the interest preference of analysis user, carries out analytical calculation, the point of interest based on user, guidance user discovery is certainly
Oneself information requirement, therefore, existing recommender system can not only provide personalized service for user, moreover it is possible between user
Establish substantial connection.
The incidence relation between data object is determined frequently with tensor resolution algorithm at present, is often adopted in recommender system
With, the algorithm of specific tensor resolution has SVD (singular value decomposition) algorithm and HOSVD (Higher-order Singular value decomposition) algorithm, still,
Often ignore the non-negative characteristic of points-scoring system currently based on the proposed algorithm of tensor resolution, that is, exists in the appraisal result obtained negative
Number, these negatives are meaningless numerical value, are usually rejected, and are impacted in this way to the effect of recommendation.
Therefore, how to avoid occurring meaningless numerical value in appraisal result, to obtain more significant numerical value to mention
High recommendation effect is a technical problem to be solved urgently.
Summary of the invention
In view of the above problems, it proposes on the present invention overcomes the above problem or at least be partially solved in order to provide one kind
State the multi-modal recommended method and device of problem.
In a first aspect, the present invention provides a kind of multi-modal recommended methods, comprising:
Attribute information based on user constructs multi-modal user's space；
Attribute information based on article constructs multi-modal article space；
Based on the multi-modal user's space and the multi-modal article space, the scoring between user and article is obtained
Amount；
Non-negative orthogonal tensor singular value decomposition is carried out to the scoring tensor, obtains the approximate tensor of the scoring tensor；
Based on the appraisal result in the approximate tensor, recommend adaptable article for user.
Further, described to be based on the multi-modal user's space and the multi-modal article space, obtain user and object
Scoring tensor between product, specifically includes:
Based on the multi-modal user's spaceThe multi-modal article spaceObtain the scoring tensor between user and article
Wherein, I1, I2..., IMThe respectively attribute information of user, J1, J2..., JNThe respectively attribute information of article.
Further, the attribute information of the user specifically include the physical condition of user, heart condition, gender, the age,
At least one of hobby；The attribute information of the article specifically include the place of production of article, size, type, on the way at least
It is a kind of.
Further, described that non-negative orthogonal tensor singular value decomposition is carried out to the scoring tensor, obtain the scoring
The approximate tensor of amount, specifically includes:
To score tensorHomomorphic Mapping obtains the matrix of low order
The matrix TM is subjected to Non-negative Matrix Factorization, specifically: the matrix TM=UM × ∑ M × VM, wherein UM >
0, VM > 0, UM be decompose after the first orthogonal matrix, VM be decompose after the second orthogonal matrix, ∑ M be decomposition after it is non-negative right
Angular moment battle array；
First orthogonal matrix UM, the second orthogonal matrix VM, non-negative diagonal matrix sigma M are intercepted respectively, corresponded to
The first low-dimensional matrix U M ', the second low-dimensional matrix V M ' and truncation diagonal matrix sigma M '；
Using homomorphism back mapping, the first low-dimensional matrix U M ' is mapped as the first non-negative characteristic tensor U ', it will be described
Second low-dimensional matrix V M ' is mapped as the second non-negative characteristic tensor V', the truncation diagonal matrix sigma M ' is mapped as non-negative diagonal
Tensor ∑ '；
By the described first non-negative characteristic tensor U ', the second non-negative characteristic tensor V', non-negative diagonal tensor ∑ ' modular multiplication is carried out,
Obtain the approximate tensor of the scoring tensor
Further, described that non-negative orthogonal tensor singular value decomposition is carried out to the scoring tensor, obtain the scoring
The approximate tensor of amount, specifically includes:
To score tensorHomomorphic Mapping obtains the matrix of low order
The matrix TM is subjected to Non-negative Matrix Factorization, specifically: the matrix TM=UM × ∑ M × VM, wherein UM >
0, VM > 0, UM be decompose after the first orthogonal matrix, VM be decompose after the second orthogonal matrix, ∑ M be decomposition after it is non-negative right
Angular moment battle array；
First orthogonal matrix UM, the second orthogonal matrix VM, non-negative diagonal matrix sigma M are intercepted respectively, corresponded to
The first low-dimensional matrix U M ', the second low-dimensional matrix V M ' and truncation diagonal matrix sigma M '；
By the first low-dimensional matrix U M ', the second low-dimensional matrix V M ', truncation diagonal matrix sigma M ' product, approximate square is obtained
Battle array
By the approximate matrix by homomorphism back mapping, the approximate tensor of the scoring tensor is obtained
Second aspect, the present invention provides a kind of multi-modal recommendation apparatus, comprising:
First building module constructs multi-modal user's space for the attribute information based on user；
Second building module constructs multi-modal article space for the attribute information based on article；
The tensor that scores obtains module, for being based on the multi-modal user's space and the multi-modal article space, obtains
Scoring tensor between user and article；
Approximate tensor obtains module, for carrying out non-negative orthogonal tensor singular value decomposition to the scoring tensor, obtains institute
The approximate tensor of the commentary amount of saying good-bye；
Recommending module, for recommending adaptable article for user based on the appraisal result in the approximate tensor.
Further, the scoring tensor obtains module, is specifically used for being based on the multi-modal user's spaceThe multi-modal article spaceObtain the scoring between user and article
Amount
Wherein, I1, I2..., IMThe respectively attribute information of user, J1, J2..., JNThe respectively attribute information of article.
Further, the attribute information of the user specifically include the physical condition of user, heart condition, gender, the age,
At least one of hobby；The attribute information of the article specifically include the place of production of article, size, type, on the way at least
It is a kind of.
The third aspect the present invention also provides a kind of computer equipment, including memory, processor and is stored in memory
Computer program that is upper and can running on a processor, the processor realize above-mentioned multi-modal recommendation side when executing described program
The step of method.
Fourth aspect, the present invention also provides a kind of computer readable storage medium, the computer readable storage medium
On be stored with computer program, the step of which realizes above-mentioned multi-modal recommended method when being executed by processor.
One or more technical solutions in the embodiment of the present invention, have at least the following technical effects or advantages:
The present invention provides a kind of multi-modal recommended methods, comprising: the attribute information based on user constructs multi-modal user
Space, the attribute information based on article construct multi-modal article space, empty based on the multi-modal user's space and multi-modal article
Between, obtain the scoring tensor between user and article；Non-negative orthogonal tensor singular value decomposition is carried out to the scoring tensor, is somebody's turn to do
The approximate tensor of scoring tensor；Based on the appraisal result in the approximation tensor, recommend adaptable article for user, solves existing
The case where with the presence of the appraisal result negative obtained in technology by proposed algorithm, influences the effect recommended, and then using non-negative
It limits, guarantees that the recommendation score result finally obtained is all nonnegative value, improve recommendation effect.
Detailed description of the invention
By reading the following detailed description of the preferred embodiment, various other advantages and benefits are common for this field
Technical staff will become clear.The drawings are only for the purpose of illustrating a preferred embodiment, and is not considered as to the present invention
Limitation.And throughout the drawings, identical component is indicated with identical reference pattern.In the accompanying drawings:
Fig. 1 shows the step flow diagram of the multi-modal recommended method in the embodiment of the present invention；
Fig. 2 shows the flow diagrams of the first tensor resolution in the embodiment of the present invention；
Fig. 3 shows the flow diagram of second of tensor resolution in the embodiment of the present invention；
Fig. 4 shows the module map of multi-modal recommendation apparatus in the embodiment of the present invention；
Fig. 5 shows a kind of structural schematic diagram of computer equipment in the embodiment of the present invention.
Specific embodiment
Exemplary embodiments of the present disclosure are described in more detail below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
It is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
It is fully disclosed to those skilled in the art.
Embodiment one
The embodiment of the present invention one provides a kind of multi-modal recommended method, as shown in Figure 1, comprising: S101, based on user's
Attribute information constructs multi-modal user's space；S102, the attribute information based on article construct multi-modal article space；S103,
Based on the multi-modal user's space and multi-modal article space, the scoring tensor between user and article is obtained；S104 comments this
The amount of saying good-bye carries out non-negative orthogonal tensor singular value decomposition, obtains the approximate tensor of the scoring tensor；S105 is based on the approximation tensor
In appraisal result, recommend adaptable article for user.
In a particular embodiment, in S101, S102, multi-modal user's space is constructed first and multi-modal article is empty
Between.Specifically, based on the attribute information of user, multi-modal user's space is constructed, wherein the attribute information of user specifically includes use
At least one of the physical condition at family, heart condition, gender, age, hobby；Attribute information based on article constructs multimode
State article space, wherein the attribute information of article specifically includes at least one of the place of production of article, size, type, purposes.
There is no successive points by S101 and S102, it can carries out simultaneously, does not do specific restriction in embodiments of the present invention.
The multidimensional characteristic that user is fully described by multi-modal customer attribute information is believed by multi-modal goods attribute
Breath fully describes the multidimensional characteristic of article.
Then, in S103, it is based on the multi-modal user's space and multi-modal article space, is obtained between user and article
Score tensor.Specifically, it is based on the multi-modal user's spaceMulti-modal article spaceObtain the scoring tensor between user and article
Wherein, I1, I2..., IMThe respectively attribute information of user, J1, J2..., JNThe respectively attribute information of article.
Due to the scoring tensorIn appraisal result be sparse, and there are null value, because
This, needs to decompose the scoring tensor, to eliminate these influences.
Especially by S104 is executed, non-negative orthogonal tensor singular value decomposition is carried out to scoring tensor, obtains the scoring tensor
Approximate tensor.
The process of specific non-negative orthogonal tensor singular value decomposition can use two ways.
The first is decomposed, as shown in Figure 2:
S201, by the tensor that scoresHomomorphic Mapping obtains the matrix of low orderThen S202 is executed.
Matrix TM is carried out Non-negative Matrix Factorization by S202, specifically: matrix TM=UM × ∑ M × VM, wherein UM > 0,
VM > 0, UM be decompose after the first orthogonal matrix, VM be decompose after the second orthogonal matrix, ∑ M be decompose after it is non-negative diagonally
Matrix；Then S203 is executed.
S203 intercepts the first orthogonal matrix UM, the second orthogonal matrix VM, non-negative diagonal matrix sigma M respectively, obtains
Corresponding first low-dimensional matrix U M ', the second low-dimensional matrix V M ' and truncation diagonal matrix sigma M '；Then S204 is executed.
First low-dimensional matrix U M ' is mapped as the first non-negative characteristic tensor U ' using homomorphism back mapping by S204, will
Second low-dimensional matrix V M ' is mapped as the second non-negative characteristic tensor V', and truncation diagonal matrix sigma M ' is mapped as non-negative diagonal
Measure ∑ '；Then S205 is executed.
S205, by the first non-negative characteristic tensor U ', the second non-negative characteristic tensor V', non-negative diagonal tensor ∑ ' carries out modular multiplication,
Obtain the approximate tensor of scoring tensor
The appraisal result that thus obtained approximation tensor shows existing Sparse when eliminating without tensor resolution,
And there is the case where null value.
Second of decomposition, as shown in Figure 3:
S301, by the tensor that scoresHomomorphic Mapping obtains the matrix of low orderThen S302 is executed.
Matrix TM is carried out Non-negative Matrix Factorization by S302, specifically: matrix TM=UM × ∑ M × VM, wherein UM > 0, VM
> 0, UM be decompose after the first orthogonal matrix, VM be decompose after the second orthogonal matrix, ∑ M be decomposition after it is non-negative to angular moment
Battle array；Then S303 is executed.
S303 intercepts the first orthogonal matrix UM, the second orthogonal matrix VM, non-negative diagonal matrix sigma M respectively, obtains
Corresponding first low-dimensional matrix U M ', the second low-dimensional matrix V M ' and truncation diagonal matrix sigma M '；Then S304 is executed.
S304 obtains the first low-dimensional matrix U M ', the second low-dimensional matrix V M ', truncation diagonal matrix sigma M ' product approximate
Matrix TM '=UM' × ∑ M' × VM'；Then S305 is executed.
S305 obtains the approximate tensor of scoring tensor by approximate matrix by homomorphism back mapping
It is first to be intercepted to the matrix obtained after decomposition in second of tensor resolution, by the matrix multiple after interception,
Approximate matrix is obtained, the approximate matrix is finally obtained into approximate tensor by homomorphism back mapping.And the first tensor resolution
In be first to be intercepted to the matrix obtained after decomposition, the matrix after interception is passed through into homomorphism back mapping respectively, is decomposed
Tensor corresponding to matrix afterwards, it is final to obtain approximate tensor by these tensors by modular multiplication.Two kinds of tensor resolutions finally obtain
The result is that identical.
After obtaining approximate tensor, S105 is executed, based on the appraisal result in the approximation tensor, recommends mutually to fit for user
The article answered.
In a particular embodiment, the appraisal result of preset threshold can will be greater than in the appraisal result in approximate tensor
Corresponding article recommends adaptable user, during recommendation, can according to appraisal result value according to by greatly to
Small sequence is recommended, wherein the article can specifically be adaptable to the physical item of user, for example, dress ornament, food etc.
Deng, can also be the news being consistent with the interest of user or advertisement etc. push text, in embodiments of the present invention just no longer
It is described in detail.
Therefore, tensor resolution and Condition of Non-Negative Constrains are avoided and is gone out in scoring tensor using technical solution of the present invention
The case where existing negative, so that the meaningful numerical value of appraisal result in scoring tensor, improves the effect of recommendation.
One or more technical solutions in the embodiment of the present invention, have at least the following technical effects or advantages:
The present invention provides a kind of multi-modal recommended methods, comprising: the attribute information based on user constructs multi-modal user
Space, the attribute information based on article construct multi-modal article space, empty based on the multi-modal user's space and multi-modal article
Between, obtain the scoring tensor between user and article；Non-negative orthogonal tensor singular value decomposition is carried out to the scoring tensor, is somebody's turn to do
The approximate tensor of scoring tensor；Based on the appraisal result in the approximation tensor, recommend adaptable article for user, solves existing
The case where with the presence of the appraisal result negative obtained in technology by proposed algorithm, influences the effect recommended, and then using non-negative
It limits, guarantees that the recommendation score result finally obtained is all nonnegative value, improve recommendation effect.
Embodiment two
Based on identical inventive concept, the embodiment of the invention also provides a kind of multi-modal recommendation apparatus, as shown in figure 4,
Include:
First building module 401 constructs multi-modal user's space for the attribute information based on user；
Second building module 402 constructs multi-modal article space for the attribute information based on article；
The tensor that scores obtains module 403, for being based on the multi-modal user's space and the multi-modal article space, obtains
Obtain the scoring tensor between user and article；
Approximate tensor obtains module 404, for carrying out non-negative orthogonal tensor singular value decomposition to the scoring tensor, obtains
The approximate tensor of the scoring tensor；
Recommending module 405, for recommending adaptable article for user based on the appraisal result in the approximate tensor.
Preferably, which obtains module 403 and is specifically used for being based on the multi-modal user's spaceThe multi-modal article spaceObtain the scoring between user and article
Amount
Wherein, I1, I2..., IMThe respectively attribute information of user, J1, J2..., JNThe respectively attribute information of article.
Preferably, the attribute information of the user specifically includes physical condition, heart condition, the gender, age, happiness of user
At least one of OK；The attribute information of the article specifically include the place of production of article, size, type, at least one on the way
Kind.
Preferably, the approximate tensor obtains module 404, specifically includes:
First map unit, for the tensor that will scoreHomomorphic Mapping obtains the matrix of low order
First decomposition unit, for the matrix TM to be carried out Non-negative Matrix Factorization, specifically: the matrix TM=UM ×
∑ M × VM, wherein UM > 0, VM > 0, UM be decompose after the first orthogonal matrix, VM be decompose after the second orthogonal matrix, ∑ M
For the non-negative diagonal matrix after decomposition；
First interception unit, for distinguishing the first orthogonal matrix UM, the second orthogonal matrix VM, non-negative diagonal matrix sigma M
It is intercepted, obtains corresponding first low-dimensional matrix U M ', the second low-dimensional matrix V M ' and truncation diagonal matrix sigma M '；
The first low-dimensional matrix U M ' is mapped as first for using homomorphism back mapping by homomorphism direction map unit
The second low-dimensional matrix V M ' is mapped as the second non-negative characteristic tensor V' by non-negative characteristic tensor U ', by the truncation to angular moment
Battle array ∑ M ' is mapped as non-negative diagonal tensor ∑ '
First approximate tensor obtaining unit, for by the described first non-negative characteristic tensor U ', the second non-negative characteristic tensor V',
Non-negative diagonal tensor ∑ ' modular multiplication is carried out, obtain the approximate tensor of the scoring tensor
Preferably, the approximate tensor obtains module 404, specifically includes:
Second map unit, for the tensor that will scoreHomomorphic Mapping obtains the matrix of low order
Second decomposition unit, for the matrix TM to be carried out Non-negative Matrix Factorization, specifically: the matrix TM=UM ×
∑ M × VM, wherein UM > 0, VM > 0, UM be decompose after the first orthogonal matrix and VM be decompose after the second orthogonal matrix, ∑ M
For the non-negative diagonal matrix after decomposition；
Second interception unit, for distinguishing the first orthogonal matrix UM, the second orthogonal matrix VM, non-negative diagonal matrix sigma M
It is intercepted, obtains corresponding first low-dimensional matrix U M ', the second low-dimensional matrix V M ' and truncation diagonal matrix sigma M '；
Approximate matrix obtaining unit is used for the first low-dimensional matrix U M ', the second low-dimensional matrix V M ', truncation to angular moment
Battle array ∑ M ' product, obtains approximate matrix TM '=UM' × ∑ M' × VM'；
Second approximate tensor obtaining unit, for the approximate matrix by homomorphism back mapping, to be obtained the scoring
The approximate tensor of tensor
Embodiment three
Based on identical inventive concept, the embodiment of the invention also provides a kind of computer equipment, including memory 504,
Processor 502 and it is stored in the computer program that can be run on memory 504 and on processor 502, the processor 502 is held
The step of realizing a kind of multi-modal recommended method when row described program.
Wherein, in Fig. 5, bus architecture (is represented) with bus 500, and bus 500 may include any number of interconnection
Bus and bridge, bus 500 will include the one or more processors represented by processor 502 and what memory 504 represented deposits
The various circuits of reservoir link together.Bus 500 can also will peripheral equipment, voltage-stablizer and management circuit etc. it
Various other circuits of class link together, and these are all it is known in the art, therefore, no longer carry out further to it herein
Description.Bus interface 506 provides interface between bus 500 and receiver 501 and transmitter 503.Receiver 501 and transmitter
503 can be the same element, i.e. transceiver, provide the unit for communicating over a transmission medium with various other devices.Place
It manages device 502 and is responsible for management bus 500 and common processing, and memory 504 can be used for storage processor 502 and execute behaviour
Used data when making.
Example IV
Based on identical inventive concept, the embodiment of the invention also provides a kind of computer readable storage medium, the meter
It is stored with computer program on calculation machine readable storage medium storing program for executing, which realizes a kind of multi-modal recommended method when being executed by processor
The step of.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications may be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (10)
1. a kind of multi-modal recommended method characterized by comprising
Attribute information based on user constructs multi-modal user's space；
Attribute information based on article constructs multi-modal article space；
Based on the multi-modal user's space and the multi-modal article space, the scoring tensor between user and article is obtained；
Non-negative orthogonal tensor singular value decomposition is carried out to the scoring tensor, obtains the approximate tensor of the scoring tensor；
Based on the appraisal result in the approximate tensor, recommend adaptable article for user.
2. the method as described in claim 1, which is characterized in that described based on the multi-modal user's space and described multi-modal
Article space obtains the scoring tensor between user and article, specifically includes:
Based on the multi-modal user's spaceThe multi-modal article space
Obtain the scoring tensor between user and article
Wherein, I1, I2..., IMThe respectively attribute information of user, J1, J2..., JNThe respectively attribute information of article.
3. method according to claim 1 or 2, which is characterized in that the attribute information of the user specifically includes the body of user
At least one of body situation, heart condition, gender, age, hobby；The attribute information of the article specifically includes the production of article
At least one of ground, size, type, purposes.
4. method according to claim 2, which is characterized in that described unusual to the non-negative orthogonal tensor of scoring tensor progress
Value is decomposed, and is obtained the approximate tensor of the scoring tensor, is specifically included:
To score tensorHomomorphic Mapping obtains the matrix of low order
The matrix TM is subjected to Non-negative Matrix Factorization, specifically: the matrix TM=UM × ∑ M × VM, wherein UM > 0, VM >
0, UM be decompose after the first orthogonal matrix, VM be decompose after the second orthogonal matrix, ∑ M be decomposition after it is non-negative to angular moment
Battle array；
First orthogonal matrix UM, the second orthogonal matrix VM, non-negative diagonal matrix sigma M are intercepted respectively, obtain corresponding the
One low-dimensional matrix U M ', the second low-dimensional matrix V M ' and truncation diagonal matrix sigma M '；
Using homomorphism back mapping, the first low-dimensional matrix U M ' is mapped as the first non-negative characteristic tensor U ', by described second
Low-dimensional matrix V M ' is mapped as the second non-negative characteristic tensor V', and the truncation diagonal matrix sigma M ' is mapped as non-negative diagonal tensor
∑'；
By the described first non-negative characteristic tensor U ', the second non-negative characteristic tensor V', non-negative diagonal tensor ∑ ' modular multiplication is carried out, it obtains
The approximate tensor of the scoring tensor
5. method according to claim 2, which is characterized in that described unusual to the non-negative orthogonal tensor of scoring tensor progress
Value is decomposed, and is obtained the approximate tensor of the scoring tensor, is specifically included:
To score tensorHomomorphic Mapping obtains the matrix of low order
The matrix TM is subjected to Non-negative Matrix Factorization, specifically: the matrix TM=UM × ∑ M × VM, wherein UM > 0, VM >
0, UM be decompose after the first orthogonal matrix, VM be decompose after the second orthogonal matrix, ∑ M be decomposition after it is non-negative to angular moment
Battle array；
First orthogonal matrix UM, the second orthogonal matrix VM, non-negative diagonal matrix sigma M are intercepted respectively, obtain corresponding the
One low-dimensional matrix U M ', the second low-dimensional matrix V M ' and truncation diagonal matrix sigma M '；
By the first low-dimensional matrix U M ', the second low-dimensional matrix V M ', truncation diagonal matrix sigma M ' product, approximate matrix is obtained
TM '=UM' × ∑ M' × VM'；
By the approximate matrix by homomorphism back mapping, the approximate tensor of the scoring tensor is obtained
6. a kind of multi-modal recommendation apparatus characterized by comprising
First building module constructs multi-modal user's space for the attribute information based on user；
Second building module constructs multi-modal article space for the attribute information based on article；
The tensor that scores obtains module, for being based on the multi-modal user's space and the multi-modal article space, obtains user
Scoring tensor between article；
Approximate tensor obtains module, for carrying out non-negative orthogonal tensor singular value decomposition to the scoring tensor, obtains institute's commentary
The approximate tensor for the amount of saying good-bye；
Recommending module, for recommending adaptable article for user based on the appraisal result in the approximate tensor.
7. device as claimed in claim 6, which is characterized in that the scoring tensor obtains module, is specifically used for based on described
Multi-modal user's spaceThe multi-modal article spaceObtain user and object
Scoring tensor between product
Wherein, I1, I2..., IMThe respectively attribute information of user, J1, J2..., JNThe respectively attribute information of article.
8. device as claimed in claim 6, which is characterized in that the attribute information of the user specifically includes the body shape of user
At least one of condition, heart condition, gender, age, hobby；The attribute information of the article specifically include article the place of production,
At least one of size, type, purposes.
9. a kind of computer equipment including memory, processor and stores the meter that can be run on a memory and on a processor
Calculation machine program, which is characterized in that the processor is realized when executing described program such as any claim institute in claim 1-5
The method and step stated.
10. a kind of computer readable storage medium, computer program, feature are stored on the computer readable storage medium
It is, which realizes method and step as claimed in any one of claims 1-5 when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811587985.7A CN109918615B (en) | 2018-12-25 | 2018-12-25 | Multi-mode recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811587985.7A CN109918615B (en) | 2018-12-25 | 2018-12-25 | Multi-mode recommendation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109918615A true CN109918615A (en) | 2019-06-21 |
| CN109918615B CN109918615B (en) | 2022-12-27 |
Family
ID=66959929
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811587985.7A Active CN109918615B (en) | 2018-12-25 | 2018-12-25 | Multi-mode recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109918615B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100185578A1 (en) * | 2009-01-22 | 2010-07-22 | Nec Laboratories America, Inc. | Social network analysis with prior knowledge and non-negative tensor factorization |
| US20120016878A1 (en) * | 2010-07-15 | 2012-01-19 | Xerox Corporation | Constrained nonnegative tensor factorization for clustering |
| JP2018128708A (en) * | 2017-02-06 | 2018-08-16 | 日本電信電話株式会社 | Tensor factor decomposition processing apparatus, tensor factor decomposition processing method and tensor factor decomposition processing program |
-
2018
- 2018-12-25 CN CN201811587985.7A patent/CN109918615B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100185578A1 (en) * | 2009-01-22 | 2010-07-22 | Nec Laboratories America, Inc. | Social network analysis with prior knowledge and non-negative tensor factorization |
| US20120016878A1 (en) * | 2010-07-15 | 2012-01-19 | Xerox Corporation | Constrained nonnegative tensor factorization for clustering |
| JP2018128708A (en) * | 2017-02-06 | 2018-08-16 | 日本電信電話株式会社 | Tensor factor decomposition processing apparatus, tensor factor decomposition processing method and tensor factor decomposition processing program |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109918615B (en) | 2022-12-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110334272B (en) | Intelligent question-answering method and device based on knowledge graph and computer storage medium | |
| CN103678431B (en) | A kind of recommendation method to be scored based on standard label and project | |
| WO2021218471A1 (en) | Neural network for image processing and related device | |
| CN109918560A (en) | A kind of answering method and device based on search engine | |
| WO2023011382A1 (en) | Recommendation method, recommendation model training method, and related product | |
| US20120109959A1 (en) | Method and system for-clustering data arising from a database | |
| US20230120054A1 (en) | Key point detection method, model training method, electronic device and storage medium | |
| CN109783749A (en) | A kind of Material for design intelligent recommendation method, apparatus and terminal device | |
| CN114912948B (en) | Cloud service-based cross-border e-commerce big data intelligent processing method, device and equipment | |
| CN110414581B (en) | Picture detection method and device, storage medium and electronic device | |
| CN111709398A (en) | Image recognition method, and training method and device of image recognition model | |
| CN114201516A (en) | User portrait construction method, information recommendation method and related device | |
| CN108960293A (en) | CTR predictor method and system based on FM algorithm | |
| CN117271818B (en) | Visual question-answering method, system, electronic equipment and storage medium | |
| CN109934631A (en) | Question and answer information processing method, device and computer equipment | |
| WO2023213157A1 (en) | Data processing method and apparatus, program product, computer device and medium | |
| CN112818213A (en) | Multimedia service data pushing method, device, equipment and storage medium | |
| CN109918615A (en) | A kind of multi-modal recommended method and device | |
| CN115827878A (en) | Statement emotion analysis method, device and equipment | |
| CN112749711B (en) | Video acquisition method and device and storage medium | |
| CN113822521A (en) | Method and device for detecting quality of question library questions and storage medium | |
| CN112862538A (en) | Method, apparatus, electronic device, and medium for predicting user preference | |
| CN112948589B (en) | Text classification method, text classification device and computer-readable storage medium | |
| CN114973333B (en) | Character interaction detection method, device, equipment and storage medium | |
| WO2023051678A1 (en) | Recommendation method and related device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |